JPS62502303A - Interconnection method of packet communication networks - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Broadcast communications network interconnection field of invention The present invention is generally formed by interconnecting multiple independent networks. separate networks without imposing any particular formal restrictions on communication systems The present invention relates to methods and configurations for interconnecting devices.

Background of the invention By interconnecting local area networks (LANs), these to form a geographically dispersed metropolitan area network This is often desired. In addition, there is a clear set of LANs within the same building. Organizations such as universities or companies that operate It becomes necessary to form a large LAN that provides services to

Devices connected to a LAN (e.g. computers, video terminals, If the device (talk device) has only one path to the devices in the LAN after interconnection, therefore If the interconnect exhibits a so-called loop-free configuration, the interconnect network by a gateway that runs the This can be achieved by connecting the work pairs. Such interconnect configurations It is called Arendt Interconnection because the existence of gateways linked together It is not visible to any equipment in the system and therefore no changes to the equipment changes are required to messages or packets propagating through That's because there isn't.

Recently, a large amount of literature has been devoted to methods and related topics for transparent interconnect configurations. Connection circuits are being discussed. These documents include (1) “Local Area Network Applications (Local Area Network Applications)”, Telecommunications.

Telecommunications, Sepetmbe r, 1984゜B. Hawe and B.Stuart (B. , Stew-art); (2) “An Architecture for Transparent Interoperability” Connection I.E.I.802 O-Cal Area Network An Archit-ecture for Transparentl y Interconnecting IEEE 802 Lo-cal Ar ea Networks)”, 0ctoter 1984 IEEE 8 Digital Equipment Corporation submitted to the 02 Standards Committee Digital Equipment Corporation Technical Paper ation Technical Paper); and (3) “Transpair Rent Interconnection of Local Networks with Bu Rydges (T r a n s p a - rent Interconnect tion of Local Networks with Bridges) ”Journal of Telecommunications Networks al of Telecommunications Networks)”.

0ctober, 1984. B, Hawe, x- -Kirby (A) and Stewart (B) ) is the author. In these papers, the physical form of the interconnection network is a branching tree. Emphasis is placed on the need to give form. The gateway forms a loop. A local area network that forms an alternative path between local networks networks cannot be interconnected in a transparent manner. In practice, the above In literature (1), a general mesh form is converted to a so-called loop-free form to generate a game. Techniques have been suggested to enable the use of

The requirements for making a system loop-free are generally strict, and ultimately limits the practical use of common gateway configurations. Satisfies channel capacity requirements In order to achieve high reliability, interconnected systems are Contains some loops in some parts. Ordinary gateways always check for these loops. Remove and remove improved redundancy or reliability. on the physical layer or link layer. The issue of loop or cyclic interconnection has not yet been discussed in the prior literature. .

Summary of the invention Local area network with storage and forwarding (store and forward) gateways What is required for loop-free configuration for transparent interconnection of networks? In the present invention, the limitations of routing information transferred by message packets are removed by methods that utilize information.

Generally speaking, the overall form of the system is to place the network at the top and the game is represented by an undirected connected line diagram with be done. A spanning tree (spanning tree) of keys for such a diagram (s panning tree) to obtain the required capacity and required redundancy.

Each spanning tree is individually identified. Each packet transmitted through the system packets to specific spanning trees so that packets are assigned to specific spanning trees. Proceed between nodes along the edges they contain. Extended storage and transmission Each gateway analyzes the packet along with the outgoing protocol and determines the assigned span. - Determine the tree and then send the message. In one embodiment of the invention, the packet The spanning tree identifier is determined by the device that sends the packet and is included in the packet. Transfer clearly or unambiguously.

Numerous unnecessary transmissions of the system due to newly connected devices to one of the networks In order to reduce the It is also possible to learn the location of the device that has been placed. In that case augmented professional The storage/delay/send algorithm is executed by the protocol.

The structure and operation of the present invention can be understood by considering the following detailed description and the accompanying drawings. be well understood.

Brief description of the drawing Figure 1 shows a pair of gateways interconnecting two local area networks. The block diagram in Figure 2 shows a block diagram of multiple local area networks with multiple gateway pairs. A loop-free three-level hierarchy system that interconnects networks. A block diagram illustrating an example, FIG. Figure 2 shows how packets propagate through the system in Figure 2 after the initial setup of the network. and demonstrate the functionality of the system to learn the location of the source of the first packet. The block diagram, Figure 4, illustrates the propagation of a return packet through the system of Figure 2. and a block indicating that the network has learned the location of the source of the reply packet. Figure 5 shows the second packet from the predetermined local area network in Figure 2. , and that the gateway has learned the identity of the source of the packet, and A block indicating that all other packets from a given device will be optimally routed. rock diagram, Figures 6 and 7 are diagrams representing the system of Figure 2 as two different spanning trees. , Figure 8 shows one cyclic and two acyclic extensions of part of the system of Figure 2. Figure 9 shows how the system shown in Figure 2 can be used to increase the system throughput. A diagram showing an example of overlapping two span trees, Figure 10 reduces traffic congestion at a single node in a part of Figure 9. A diagram showing an example of providing a duplicate node to Figure 11 shows an example of the steps of packet propagation through a general cyclic system. This is a flowchart.

detailed description To make the explanation clear, we will first explain the ordinary gateway configuration with a focus on methodology. It is useful to provide a basis for the basic concept of the invention. on this basis Basic concepts that make it easier to later present embodiments starting from the prior art and terms are introduced.

1. Basic matters In FIG. 1, the letters NETWORKA and NETlt! Two with 0RKB The networks 101 and 102 are connected to a pair of unidirectional gateways 102 and 202. The entire communication system 100 is formed by interconnecting the communication system 100 with each other. gateway 10 2 and 202, the gate 202 is connected to the network NETWOR by the gate 102. The link is connected in such a manner that transmission to the KB is ignored. Gate 102 is also a gate 202, the transmission to the network NETWORKA is ignored.

Each individual network typically provides sending and receiving services to multiple devices or hosts. I do. In Figure 1, hosts 111 and 112 and hosts 211 and 212 are It connects to networks 101 and 201, respectively, and communicates via these networks. practice faith. Each host shall have a unique identification address, and each The network has a broadcast communication function, that is, if one host sends a packet, it All other gateways in the network will receive this packet.

Each packet transferred on the network has a sending host address and a destination host address. Includes dress and actual message date. interconnecting two networks One of the gateways receives each packet sent to a given network. Arrange the configuration as follows. For example, in FIG. 1, gateway 102 is network 1 is the receiving gateway for all packets sent on 01.

The storage and transmission protocols implemented by each gateway are as follows.

In other words, the gateway receives a packet addressed to the host that previously appeared in the sending address of another packet. packets, and sends each packet to be transferred on the corresponding network. No. If host 1 first sends a packet, each gateway learn the location of the host. Later the second host sends a packet to the first host. When a packet is sent, the packet automatically follows the optimal path through the communication system.

To explain this protocol in detail, consider communication system 300 of FIG.

This configuration consists of a gateway pair consisting of gateways that execute the above algorithm. Contains many networks connected through.

The following symbols are used to specify the protocol precisely.

NI: Network number tone (element 30 i), i=1゜---,? G, ,: Gateway from network i to network j (element 3ij) A, B, C, D, E: Host computer (elements 381 and 3, respectively) 82. ---. 385) P xv: Packet sent from host X to host Y is “retransmitting” DIj: Transmission via GIJ is blocked and hosts accumulated in GiJ are (DIJ is called a "drop filter").

The configuration in Figure 2 is initially set as Di, = φ (empty set) for all i, j. Assume that a transmission (PE) is performed from host (PE) E to host C (PE).

1, PEC is transmitted through N7.

2, D, 3=(E), and P. 3 to N3 via Gt3, Das=(E ), and P EC'G 3+ to N, 4, D, 2 = (E), and PEC 5 to N2 via G12, D2. = (E) and D to N4 via PECG24 2S=(E), and PEC02% to N, PEC in step 3 above. reaches its destination, but continues on its way through the network. This is the broken line in Figure 3. Indicated by

Next, consider the return packet Pet with reference to FIG.

1, PCE is transmitted N6.

2, D63=(C), and 3, D3. to N3 via PCE cs3. = (C ,E), and Pet are dropped via G31, =(C), and Bipet Ga? via Nt (and host E) Since the return packet is “known” from the source, the return packet will take the optimal route through the network. pass through.

As a final consideration, let us consider the second bucket) PEC as shown in Figure 5. Assuming you need to send it through.

1, PEC is transmitted through N7.

2.D? 3=(E), and N via PECG24 to 3, D3. =<c, E), and Pie is dropped via G31 = (E), and PkE per 6E to N (and host C) via PECG36 to virtually any is transmitted only through the necessary networks. P CE is G s3+ Gs+ and G 3? Since only a portion of the entire network is received by host C's location, I just know that. but the rest of the network sends it to other hosts If so, learn about host C. When steady state is reached, each Gi, is optimal So-called “blocks” or “drops” that indicate routing and prohibit extra transmissions. It has a list DIJ.

2. Extensions to protocol configuration As is clear from the above description, the system 300 allows packets to be routed through two gates. one between any pair of hosts to prevent perpetual looping between hosts. By placing the gateway pair on the edge (side), If the system is placed on a diagram with no specified direction, the resulting diagram will be Because of the movement, it is necessary to avoid forming loops. is restricted to the form of an acyclic diagram. FIG. 6 is a tree diagram showing the configuration of the system 300. It is shown in the form of The gateway pairs shown in Figure 2 are grouped together and shown in Figure 6. By writing in the numbers, the edge or side (solid line) is defined. this tree The diagram is acyclic because each pair of nodes is connected by only one path. Ru. For example, if nodes 306 and 307 are directly connected (dashed line in Figure 6) As shown by the edge 375 (<), this tree diagram is in cyclic form.

Circular systems directly implement the storage and output protocols or algorithms described above. cannot be used for For example, from FIG. 6, nodes 306 and 307 are also directly connected. is sent from the host at node 303 to the host at node 306. The received packet is sent to the path 303-307-306-303-307--- 303-306. A looping state occurs at -307-303-306---. looping shape packets saturating the gateway and network in its loop. This will prevent normal communication.

The structure shown by the solid line in Figure 6 is a spanning tree (spanning tree). ng tree)), that is, in span-free, each node pair has one connected only by passageways.

The tree diagram of FIG. 7 shows an alternative spanning tree for the node configuration of FIG.

Generally, any tree diagram has multiple spanning trees.

In order to improve the gateway-protocol configuration according to the present invention, certain guidelines are followed. A set of spans 71J- is selected for the cyclic graph. Each spanning tree has a special A specific identifier or number is assigned to each message transmitted through the system. A particular spanning tree is assigned via its identifier. Receive this message Determine the tree number for whichever gateway you want to use, and then send a message to identify span tree and drop all packets in other span trees. make it drop. Typically, the device that sends the message The number is clearly or unambiguously identified. For example, if it is clearly specified, A “tree number field” can be added to the packet specification, i.e. Can be added as additional bits to the header. If it is unclearly specified , span tree numbers are commonly used in packets such as packet source and destination addresses. can be generated from commonly occurring fields. An example of a suitable function is Span tree number = (packet source ■ destination) modulo N, where N is the network is the number of spanning trees in the network. This is everything between a pair of hosts. of traffic is carried over only one spanning tree, and therefore the system This has the advantage of minimizing the droplists used in the process.

The tree diagram in the center of FIG. 8 corresponds to node 303 in the tree diagram of FIG. , 306 and 307 and solid and dashed lines including edges 364, 374 and 375 This is a reconstruction of the parts shown in both. The tree diagram in Figure 8 is cyclic. Ru. To arrive at this new configuration, for example, the system 300 of FIG. One gateway pair (edge 375) connecting nodes 306 and 307 Assume that it has been set up. Message from node 306 to 307 by edge 375 Since the traffic can be increased, Edge 364 and 374 are no longer overloaded. It becomes impossible to process it.

Since the tree diagram of FIG. 8 is cyclic, - sets of span trees are selected.

Two span trees covering this tree diagram are drawn on the left and right sides of Figure 8. They are shown as FIGS. 401 and 402, respectively. When using the symbolic representation M(S;D;T) (where S is the packet source node, D is the destination node set, and T is the span tree number. ), one possible message route for nodes 303, 306 and 307 The algorithm for allocation is as follows.

M (303; 306, 307, 301, ---; 2).

M (306; 303.307.301.---; 1), and M (30 7; 303, 306, 301,, ---; 2>.

This particular allocation is from node 306, presumably for load balancing purposes. Only edge 375 is used for messages that occur. Edge 375 is inoperable , the device associated with node 306 is M (306; 303, 307° 30 1, ---; 2) is notified of the assignment change and thus maintains communication within the system. can. Multiple spanning trees in a cyclic network allow the corresponding gates to There will be some loss of operation during periods of way inactivity, but some nodes will have redundancy. given.

In general, gateways that appear in different spanning trees are You need to maintain a drop list. For example, from Figure 8, the number span ・If the drop list for the tree is expressed as D (k), then the drop list on both trees is A packet (PC,) sent from host C to host E is sent to edge 374. At startup, D(1)3-=φ, D<2), F=(C).

D(1)? 3=(C) and D(2)ts=φ. multiple droplists In any tree diagram, the need to maintain the majority of the spanning tree, Select in tuples such that an edge is not contained in more than one spanning tree It can be reduced by doing.

Even if the system is acyclic, it may be sufficient in practice to use a multi-spanning tree. It is often necessary to obtain sufficient communication capacity. The present invention is in this state. can also be dealt with. Examples of such cases include public or private telephone network There are tree-like systems, such as Araki, that cover large areas. Such a system Mums exhibit a tendency to become bottlenecks at or near the roots. one edge Use unconnected spanning trees that are not included in more than one spanning tree. Instead of considering capacity, you can weight approximately the same spanning trees on the tree diagram. You will need to sleep. This is shown in the diagram in Figure 9, which converts Figure 6 into a three-level diagram. Rewritten to show two spanning trees per hierarchy (one (one is shown with a solid line and the other with a dashed line).

In this case, edge 343 is one gateway in each of the spanning trees. Realized only with pairs. However, the edge 322 is one for each spanned tree. Implemented with two gateway pairs. Similarly, edges 353, 364 and 374 are Implemented with one gateway pair for both trees, while edge 332 has two It has two gateway pairs. Because of the configuration using this method, Edge 343 (same as The drop list for each span tool (353, 364 and 374) droplist, thus requiring only one droplist to be maintained. It's just a matter of doing.

Similarly, the drop list for Edge 322 (also Edge 332) will be the same. However, this is less important since two separate gateways are used.

The network in Figure 9 is a gate between levels 1 and 2 (edges 322 and 332). Ways are duplicated and placed to reduce conflicts. However, comparison is also possible at nodes. It can happen. FIG. 10 shows a replication of node 301 in a portion of the network of FIG. A node (node 308) is provided to reduce traffic passing through node 301 An example is shown below. This network has hosts connected to nodes 301 and 308. It will operate properly unless it is connected. Hosts are connected to nodes 301 and 308 In some cases, add a gateway pair to Edge 384 and 394 for complete connectivity. It is necessary to have the following.

A way to add parallel edges and nodes throughout the system is to place duplicates. There is no limit to the number of levels or the number of parallel edges and nodes added. mosquito Due to these characteristics, networks with arbitrarily large throughputs can be constructed. . Such networks can be replicated independently of the number of spanning trees used. The desirable feature is that a drop edge can be realized with only a single drop filter. have sex.

Figure 11 Propagates Each Packet Through a System Using Spanning Trees 1 shows a flowchart of an example of the steps for. Block 401 includes capacity and reliability. The set of spanning trees for the system is designed to satisfy reliability requirements. The tree is first selected and then each tree is assigned a unique identifier. Bro As shown in block 411, the packet is generated and then transmitted over its associated source network. For each packet propagated by a device, a preselected The span tree identifier that has been entered is included in the packet. Receive propagated packets Each gateway transmits the identifier contained in the packet and the packet that generated the packet. Determine the source device and the destination device that will receive the packet, and do this in block 421. Indicated by As indicated by block 422, the address of the packet source is still If it does not exist, this address will inserted into the list. Also, as shown by decision block 423, the gate if the method is not processing a particular spanning tree that can be found in the packet. , this packet is dropped. Additionally, as shown in decision block 425, the packet If the destination of the drop is found in the drop list for the spanned tree, If so, the packet is dropped; otherwise, as shown in block 426 The packets are broadcast by the gateway. Packet detected at destination device and typically a confirmation packet is sent to the packet source device, as shown in block 431. It will be sent back.

The return packet does not necessarily have to follow the same spanning tree as the original packet. For efficiency reasons, it is necessary to follow the same tree as much as possible.

On any system, especially large systems, when a new host is brought into service, Until the system learns the location of the new host, Messages from a network to a new host may be flooded. Regaru. This point is also another aspect of the present invention, and to illustrate this point, the right side of FIG. That is, networks 303, 306 and 307 (Na, N6 and N,), Gateways 337 and 373 (G,? and G,3) and hosts 384 (D) and and 385(E). host 384 and network 307 Assume that there is a packet p■ that is transmitted. Gateway G73 is a host 384, so the drop filter is not in that list. Since we do not have host D, this packet would normally spill out of the system. deer and gate G7. is longer than the average recognition response time of the host 384 ('rn ), gateway G73 will Add host 384 to its droplist (D, 3 = (E, D, ---)') Add. In reconsidering the packet PED by gateway Gy3, the system This packet propagates through and retransmits to network 303(N,). Prevented. This means that the host 384 is either a telephone or a data set. Generalized to describe communication devices and networks 306.307. -1- This is a particularly important consideration if the network is a locally switched network.

This pre-send delay protocol can also be combined with the spanned tree protocol to achieve the present invention. can be made to form other aspects of the . Block 42 in FIG. 5 and 426, through an accumulation and delay block followed by another decision block. Can be inserted. After a predetermined delay in the first new block, block 425 If a new decision block is set, a new decision block is tested for the same conditions. destination address If the response is in the droplist, the packet is dropped, otherwise the packet is dropped. The packet is processed by block 426.

Other aspects of the invention reduce network overflow and reduce delays. Therefore, a pair of gateways can be configured to communicate their drop lists with each other. subordinate Therefore, host D has a gateway indicating that host D is not on network 307. Hey G3? Drop list for (D,, (---.

D, ---)), the FED must be repeated immediately. Ru. (bucket) PE, combining both communication and delay. Gateway G for 73 and G2? The complete algorithm for is as follows.

1.D is D? 3, the packet l”Pto is not repeated; 2. If D is included in D37, packet PEON3; 3. D is included in D3□ If Zukatsu D is not included in D73, bucket) Ptt+ is G? 3 at time T It is delayed or stored for fl and then fed again.

For gateway pairs that introduce time delays, block 425 of FIG. If the target device does not obtain a positive judgment result in the forward direction, the destination device uses the back propagation method. test to determine if it exists in the droplist for the tree in the direction Can be extended to run If the result of this additional test is positive, the packet propagate onto the associated spanning tree. Additional test results are negative If , the packets are accumulated over a clearly selected time interval and the next is fed back to block 425 to obtain the final result.

The methods and associated apparatus described above are not meant to limit the invention. However, other embodiments are limited only by the scope of the claims of the present invention. You can also

FIG.5 FIG.B international search report +, lIeMsaaw+A ue+++: *n1laPCT/US861012 06ANNEX To THE INTERNATZONAI:, 5EARC HREPORT ON-m------1-++-----phrase---groan---- ・−1・−1−−Cloudy−−−−Yu−−

Claims (8)

[Claims] 1. Multiple networks interconnected by gateways that perform drop list processing In a packet transmission method in a system including a network, Select a set of spanning trees for the system and create an identifier representing one of the trees. transmitting Besshi by the packet; the packet being routed through the system by the gateway according to the identifier; 1. A packet transmission method comprising the step of performing routing. 2. The packet transmission method according to claim 1, wherein the one of the trees is A method for transmitting a packet including the step of retransmitting an authentication packet. 3. For systems containing multiple networks interconnected by gateways. connecting said packet to one of said networks for transmitting said packet; packets originating from the source device and connected to one of said networks. In the method of transmitting packets to the destination device, Select a spanning tree for the system and assign the tree to each of the gateways. form a drop list for transmitting along with the packet an identifier indicating one of the trees; The packet is forwarded by the packet source device on the tree corresponding to the identifier. For each gateway that broadcasts through the system and receives said packets ( i) packet source address, destination address and span, tree for said packet; - determining an identifier; (ii) determining a droplist associated with said spanning tree identifier; (iii) the gateway is identified; If the spanned tree is not processed, the packet will be dropped, otherwise the packet will be dropped. If (iv) the destination address is If present in the list, cause said packet to be dropped; otherwise, said packet is send a packet onto the identified tree A bucket transmission method that includes steps. 4. In the packet transmission method according to claim 3, on the identified tree: The destination device receives the packet by broadcasting a return packet. A method for transmitting a packet including the step of confirming receipt of the packet. 5. In the packet transmission method according to claim 3, before the result of the tree selection If the same drop list as mentioned above occurs, it will be shared with any of the above gateways. A method for transmitting packets comprising the step of maintaining a drop list of packets. 6. In the packet transmission method according to claim 3, the substep (iii) next, If the destination address is in the drop list, drop the packet. If not, accumulate said packets over a predetermined time interval and then and transmitting said packet for processing according to substep (iv). How to transmit packets containing. 7. A packet that propagates through a gateway that performs drop list processing. In the packet propagation method, the packet source address is transmitted from the packet within the gateway. extract the response and destination address, If the packet source address is not in the drop list, the drop add the packet source address to the pre-list; If the destination address is not in the drop list, drop the packet. accumulate over a fixed time interval, during which subsequent packets are forwarded by the gateway. process, If the destination address is not in the drop list after the predetermined time interval; If so, send said packet, otherwise drop said packet. A packet propagation method characterized by including steps. 8. A method of processing packets by a first gateway from a pair of gateways. Thus, the first gateway of the gateway pair is read by the second gateway. and the second gateway has a first drop list that can be accessed by the first gateway. packet processing method having a second drop list readable by the network; and the packet source address and destination address are determined from the packet within the first gateway. extract the destination address and the packet source address is in the first drop list. if not, inserting the packet source address into the first drop list; If the destination address is not in the second drop list, the packet is sent by the first gateway; If the packet source address is not originally in the first drop list and If the specified destination address is not in the second drop list, the packet is dropped. accumulated over a predetermined time interval by said first gateway, at which time subsequent packets are is processed by said gateway pair, the destination address is not present in the first drop list after the predetermined time interval; If yes, send said packet, otherwise drop said packet. A packet processing method comprising the step of: